afr: Block fops when file is in split-brain

For directories, block metadata FOPS.
For non-directories, block data and metadata FOPS.
Do not block entry FOPS.

Change-Id: Id7f656f4a513b9d33c457dd7f2d58028dbef8e61
BUG: 1235007
Signed-off-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-on: http://review.gluster.org/11371
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Tested-by: NetBSD Build System <jenkins@build.gluster.org>

4 files changed, 76 insertions, 14 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index f644c9dc200..b42772fda09 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -336,6 +336,58 @@ out:
 }
 
 int
+afr_inode_get_readable (call_frame_t *frame, inode_t *inode, xlator_t *this,
+                       unsigned char *readable, int *event_p, int type)
+{
+
+        afr_private_t *priv = this->private;
+        afr_local_t *local = frame->local;
+        unsigned char *data = alloca0 (priv->child_count);
+        unsigned char *metadata = alloca0 (priv->child_count);
+        int data_count = 0;
+        int metadata_count = 0;
+        int event_generation = 0;
+        int ret = 0;
+
+        /* We don't care about split-brains for entry transactions. */
+        if (type == AFR_ENTRY_TRANSACTION || type == AFR_ENTRY_RENAME_TRANSACTION)
+                return 0;
+
+        ret = afr_inode_read_subvol_get (inode, this, data, metadata,
+                                         &event_generation);
+        if (ret == -1)
+                return -EIO;
+
+        data_count = AFR_COUNT (data, priv->child_count);
+        metadata_count = AFR_COUNT (metadata, priv->child_count);
+
+        if (inode->ia_type == IA_IFDIR) {
+                /* For directories, allow even if it is in data split-brain. */
+                if (type == AFR_METADATA_TRANSACTION) {
+                        if (!metadata_count)
+                                return -EIO;
+                }
+        } else {
+                /* For files, abort in case of data/metadata split-brain. */
+                if (!data_count || !metadata_count)
+                        return -EIO;
+        }
+
+        if (type == AFR_METADATA_TRANSACTION && readable)
+                memcpy (readable, metadata, priv->child_count * sizeof *metadata);
+        if (type == AFR_DATA_TRANSACTION && readable) {
+                if (!data_count)
+                        memcpy (readable, local->child_up,
+                                priv->child_count * sizeof *readable);
+                else
+                        memcpy (readable, data, priv->child_count * sizeof *data);
+        }
+        if (event_p)
+                *event_p = event_generation;
+        return 0;
+}
+
+int
 afr_inode_split_brain_choice_get (inode_t *inode, xlator_t *this,
                                   int *spb_choice)
 {
@@ -593,6 +645,8 @@ afr_accuse_smallfiles (xlator_t *this, struct afr_reply *replies,
 	for (i = 0; i < priv->child_count; i++) {
 		if (data_accused[i])
 			continue;
+                if ((priv->arbiter_count == 1) && (i == ARBITER_BRICK_INDEX))
+                        continue;
 		if (replies[i].poststat.ia_size < maxsize)
 			data_accused[i] = 1;
 	}
@@ -1677,6 +1731,10 @@ afr_local_discovery_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
          * the slowest local subvolume is far preferable to a remote one.
          */
         if (is_local) {
+                /* Don't set arbiter as read child. */
+                if ((priv->arbiter_count == 1) &&
+                    (child_index == ARBITER_BRICK_INDEX))
+                        goto out;
                 gf_log (this->name, GF_LOG_INFO,
                         "selecting local read_child %s",
                         priv->children[child_index]->name);
diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c
index 6121108872f..6e545497108 100644
--- a/xlators/cluster/afr/src/afr-read-txn.c
+++ b/xlators/cluster/afr/src/afr-read-txn.c
@@ -52,6 +52,9 @@ afr_read_txn_next_subvol (call_frame_t *frame, xlator_t *this)
                 local->op_ret = ret;                              \
                 local->op_errno = errnum;                          \
                 read_subvol = index;                              \
+                gf_msg (this->name, GF_LOG_ERROR, EIO, AFR_MSG_SPLIT_BRAIN,\
+                        "Failing %s on gfid %s: split-brain observed.",\
+                        gf_fop_list[local->op], uuid_utoa (inode->gfid));\
                 goto label;                                       \
         } while (0)
 
@@ -59,7 +62,6 @@ int
 afr_read_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err)
 {
 	afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
 	int read_subvol = 0;
 	int event_generation = 0;
 	inode_t *inode = NULL;
@@ -68,27 +70,19 @@ afr_read_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err)
 
 	local = frame->local;
 	inode = local->inode;
-        priv  = frame->this->private;
 
 	if (err)
                 AFR_READ_TXN_SET_ERROR_AND_GOTO (-1, -err, -1, readfn);
 
-	ret = afr_inode_read_subvol_type_get (inode, this, local->readable,
-					      &event_generation,
-					      local->transaction.type);
+	ret = afr_inode_get_readable (frame, inode, this, local->readable,
+			              &event_generation,
+				      local->transaction.type);
 
 	if (ret == -1 || !event_generation)
 		/* Even after refresh, we don't have a good
 		   read subvolume. Time to bail */
                 AFR_READ_TXN_SET_ERROR_AND_GOTO (-1, EIO, -1, readfn);
 
-         /* For directories in split-brain, we need to allow all fops
-          * except (f)getxattr and access. */
-        if (!AFR_COUNT(local->readable, priv->child_count) &&
-            local->transaction.type == AFR_DATA_TRANSACTION &&
-            inode->ia_type == IA_IFDIR)
-                memcpy (local->readable, local->child_up, priv->child_count);
-
 	read_subvol = afr_read_subvol_select_by_policy (inode, this,
 							local->readable, NULL);
 	if (read_subvol == -1)
@@ -237,8 +231,8 @@ afr_read_txn (call_frame_t *frame, xlator_t *this, inode_t *inode,
 	if (read_subvol < 0 || read_subvol > priv->child_count) {
 		gf_msg (this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN,
                        "Unreadable subvolume %d found with event generation "
-                       "%d. (Possible split-brain)",
-                        read_subvol, event_generation);
+                       "%d for gfid %s. (Possible split-brain)",
+                        read_subvol, event_generation, uuid_utoa(inode->gfid));
 		goto refresh;
 	}
 
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index a2023884465..b27cfedaddb 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -1967,6 +1967,13 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
         if (ret < 0)
             goto out;
 
+        ret = afr_inode_get_readable (frame, local->inode, this, 0, 0, type);
+        if (ret) {
+                gf_msg (this->name, GF_LOG_ERROR, EIO, AFR_MSG_SPLIT_BRAIN,
+                        "Failing %s on gfid %s: split-brain observed.",
+                        gf_fop_list[local->op], uuid_utoa (local->inode->gfid));
+                goto out;
+        }
         afr_transaction_eager_lock_init (local, this);
 
         if (local->fd && local->transaction.eager_lock_on)
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 0bb184c78ae..c8e1a5f0008 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -761,6 +761,9 @@ typedef struct afr_read_subvol_args {
                                             (op_errno == EBADFD)))
 
 int
+afr_inode_get_readable (call_frame_t *frame, inode_t *inode, xlator_t *this,
+                        unsigned char *readable, int *event_p, int type);
+int
 afr_inode_read_subvol_get (inode_t *inode, xlator_t *this,
 			   unsigned char *data_subvols,
 			   unsigned char *metadata_subvols,